Microsoft has introduced its latest innovation in AI technology with the launch of Phi-3-mini for smartphones. It is amongst the first three small models the company plans on releasing.
In a market where Large Language Models (LLMs) have opened up exciting new opportunities for enhanced productivity and creativity through AI, it is crucial to acknowledge that their sheer size demands significant computing power, which is not feasible for everyone. On the other side, the Phi-3 models emerge as the most efficient and budget-friendly Small Language Models (SLMs) available. Despite their compact nature, these SLMs boast the same functionalities as their larger counterparts. They’re trained on smaller datasets yet deliver impressive performance, particularly for less complex tasks. This makes them not only more accessible but also simpler to integrate for organizations operating with limited resources. Moreover, their flexibility allows for easier customization to suit specific requirements.
Sonali Yadav, Microsoft’s principal product manager, highlighted the transition towards smaller-scale models. She emphasized that these models offer users the flexibility to select the most suitable one for their specific needs and situations, ushering in a new era of personalized AI experiences.
The Phi-3-mini comes with a 3.8 billion parameters, making it a powerhouse in its class. Unlike larger language models such as GPT-4, it’s trained on a relatively smaller dataset. Parameters essentially signify the number of intricate commands a model can comprehend, highlighting its capability to handle a wide range of tasks.
Having a variety of models allows for selecting the most appropriate one for a given device. Smaller models such as Phi-3-mini are more economical to operate and deliver superior performance on smartphones and laptops compared to their larger counterparts. Sébastien Bubeck, Microsoft’s vice president of GenAI research, emphasized that Phi-3 doesn’t just offer a slight cost advantage; it presents a significant cost-saving opportunity, boasting a tenfold difference in expenses compared to other models with similar capabilities.
During a chat with The Verge, Eric Boyd, who is the corporate vice president of Microsoft Azure AI Platform, explained how the latest AI model differs from its predecessor. He mentioned that Phi-3 builds upon the foundations laid by its predecessors. While Phi-1 primarily focused on coding, and Phi-2 started grasping reasoning, Phi-3 excels in both coding and reasoning abilities. Even though the Phi-3 model family possesses some general knowledge, it still falls short compared to a GPT-4 or another large language model (LLM) in terms of breadth. There’s a substantial contrast in the quality of responses between an LLM trained on the entire internet and a smaller model like Phi-3.
In today’s AI-driven world, every tech giant boasts its own AI innovation, from Google’s Gemini to OpenAI’s ChatGPT. As technology continues to advance, the emergence of smaller and more accessible AI models like the Phi 3-mini is becoming increasingly inevitable. This compact yet powerful model is now readily accessible through various platforms, including Microsoft’s cloud service platform Azure’s AI model catalog, the machine learning model platform Hugging Face, and Ollama, a framework designed for running models on local machines.
